变分自编码模型在周期性KPI指标异常检测中的应用研究

Research on the application of VAE model in periodic KPI detection

随着5G时代的来临,用户对网络感知的要求不断提升,相应地对网络的感知优化的工作也提出了更高要求。为了更加精准、及时的定位网络内的异常,将VAE(Variational Auto-Encoder,变分自编码)模型应用于质差小区周期性KPI(Key Performance Indicator,关键绩效指标)指标异常的检测。通过对周期性数据编码、解码后还原KPI指标分布,分析其与实际值的差距,区分出周期性KPI指标中的异常和噪声,最终提升质差小区优化的及时性和有效性。

With the coming of 5 G Era, user’s requirements of network quality are constantly increasing. Correspondingly, there are higher requirements for the work of network perception optimization. In order to locate the anomalies in the network more accurately and timely, applying the VAE model in detecting the periodic KPI in the quality poor cell. By coding and decoding the periodic data and restoring the KPI distribution, the gap between the analysis and the actual value is analyzed, and the anomalies and fluctuations in the periodic KPI are distinguished. Finally, Improve the timeliness and effectiveness of poor quality cell optimization.